Links: AI planning resources on the Web

Abstract

W ith the introduction of new, efficient techniques for planning, interest in the field has risen sharply in recent years. Planning systems have come a long way since the days of Blocks World. Today, planning and scheduling techniques are being used to solve problems in military campaign planning, robot navigation, industrial equipment scheduling, human-computer interaction, and many other real-world domains. Planning, which is the process of finding a sequence of actions that meets an agents goal, is a thriving area of artificial intelligence (AI) research and is growing almost faster than one can keep track. Fortunately, an abundance of information about planning is available, in both paper and electronic forms; answers to most of the questions a newcomer to the field might ask can easily be found. Early approaches to planning , as exemplified by STRIPS (Stanford Research Institute Problem Solver), viewed the planning problem as that of generating a sequence of operators that will transform the current state of the environment into a goal state. This approach is sometimes referred to in the planning literature as " state space search. " Eventually a different perspective gained dominance , in which states represented not properties of the environment but the plans being considered. That is, rather than searching through a space of world states, planning systems searched through a space of partially elaborated plans. Penberthy and Weld's UCPOP, a partial-order planner, is one of the best known examples of this approach. More recently, with the development of the Graphplan algorithm by Blum and Furst, another conceptual shift has taken place. Many systems now treat planning as a form of constraint satisfaction, exploiting new work on efficient algorithms in this area. These are just a few high points in a rich and very branchy history of planning systems. A good historical introduction to planning starts with an article in AI Magazine titled " AI Planning: Systems and Techniques " (Hendler et al. 1990). Beginning with a description of the planning problem as " designing systems which can describe a set of actions (or plan) which can be expected to allow the system to reach a desired goal, " Hendler et al. discuss common techniques used for planning, a chronology of planning systems, and some of the problems that planning addresses: reasoning about time, physical constraints on solutions, execution uncertainty, perception, and multi-agent systems. In addition to Hendler et al.'s discussion of …